What is the Jev-Lite decision engine?

An open-source, non-autoregressive System One decision engine inspired by TypeSafe AI. It evaluates state context against typed questions in single-pass forward tensor operations and returns probability distributions and calibrated confidence scores.

How is it different from calling an LLM?

The README's argument: LLMs are System Two, and forcing them to emit routing decisions as JSON costs latency and tokens, can hallucinate fields and breaks parsing. Jev-Lite skips generation entirely, so there is no generated text to parse.

What powers it under the hood?

Per the README badges: Python 3.11 or 3.12, FastAPI 0.115+, PyTorch 2.2+, and a Qwen 2.5 0.5B or 3B backbone, with decision-engine, deep-learning and zero-shot among the listed topics.

Is it a hosted service or something I run?

The repository presents it as an engine you install and run yourself, built around FastAPI and PyTorch; no hosted endpoint or API key is described in the captured text.

What does it cost to run Jev-Lite?

Nothing is priced in the captured README: there is no hosted tier or API key described, just the Apache-2.0 repository you run on your own hardware. GPU requirements are not stated in the excerpt.

What questions can it answer?

Typed questions over state context, in the Jev mold: the README's introduction names routing decisions and classifications as the target use, returning probability distributions rather than text.

Who built it and how active is it?

It is Atulmishra22's Apache-2.0 Python repository, created 2026-09-29 and last pushed 2026-10-04, with no stars or forks; the README shows no benchmarks or release history.

Jev directory

CommunityJev Repos & SDKs

Jev-Lite

An open-source, non-autoregressive System One decision engine inspired by TypeSafe AI: it evaluates state context against typed questions in a single forward pass, returning probability distributions and calibrated confidence without generating text.

See highlights Open github.com

Stars: 0
Language: Python
License: APACHE-2.0
Reading time: 2 min
Published by: Community
Author: Atulmishra22
Last verified: Oct 2026

Atulmishra22/jev-lite

User repository on GitHub

View on GitHub

Production-grade System One decision engine inspired by TypeSafe AI. Evaluates state and typed questions in a single forward pass with zero generation latency.

Forks: 0
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

Tagscommunity github python pytorch fastapi

Works forJev for Classification Jev for Routing Jev for Scoring

Used inJev for Software & DevTools

Languages: Python
Platforms: API
Capabilities: Open sourceSelf-hostedRuns locally

Highlights

Single-pass forward tensor operations replace token generation, removing generation latency and JSON parsing fragility.
Returns mathematical probability distributions and calibrated confidence scores for typed questions.
Stack: Python 3.11 or 3.12, FastAPI 0.115+, PyTorch 2.2+, on a Qwen 2.5 0.5B or 3B backbone.
Apache-2.0 licensed and framed as production-grade; the README argues JSON prompting a chatbot adds latency, token cost and hallucinated fields.
The README describes an open-source engine you run yourself rather than a hosted API.

Watch out

README reviewed 2026-10-17; the captured text stops mid-sentence and the repository publishes no benchmarks, accuracy or memory figures, so the production-grade claim is the project's framing. It is unrelated to the vagmi/jev-lite Hugging Face adapter already listed in this directory.

Questions about this entry

Short answers drawn from the linked project or page.

From the Jev guides

Original write-ups on what this entry is for.

The Open Reimplementations: Local Models That Speak System One

Sep 22, 2026

Closed weights have not stopped the ecosystem from rebuilding the shape in the open. API-compatible servers, retrained heads on small models, and inference-time tricks that skip generation entirely.

Jev Repos & SDKs#open-models#research#local

Read article

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai